<p>Geographical origin affects the quality traits and market value of Citri Reticulatae Pericarpium. Thus, a rapid, nondestructive, and reliable origin identification method is needed. Visible and near infrared spectroscopy offers fast detection, simple sample preparation, and rich spectral information. However, traditional machine learning methods show limited spectral representation ability, while deep learning models can be influenced by local variations. To address this issue, this study proposed an origin identification method based on an attention-enhanced ResNet. Visible and near infrared spectra from samples of different origins were collected, and preprocessing methods were used to improve spectral quality. A one-dimensional residual network was then developed for classification, and an attention mechanism was introduced to enhance key spectral responses and suppress irrelevant information. The network structure was further optimized within the residual learning framework to improve multiscale feature representation. The proposed method was evaluated through comparative and ablation experiments. Results showed that it could better capture spectral differences related to origin and achieved higher accuracy, stability, and generalization ability. This study provides a new approach for the nondestructive origin identification of Citri Reticulatae Pericarpium and a useful reference for intelligent spectral identification of medicinal and edible materials.</p>